CT Image Noise Estimation Using Total Attenuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CT image noise reduction methods introduce artifacts like ringing near sharp transitions and fail to adapt effectively to varying noise levels across different slices due to anatomical changes, particularly in head scans with highly attenuating bones.
Innovation Solution
The method estimates noise levels using the total attenuation of reconstructed images and applies a multi-scale anisotropic diffusion filtering technique, followed by an enhancement filter to sharpen edges and adaptively adjust processing based on empirical noise estimation, utilizing wavelet transforms for efficient noise removal across different scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional wavelet based noise suppression techniques are applied, then noise reduction is achieved, but artifacts such as ringing are introduced near sharp transitions in the image
Solution Approach 1:
The patent applies different processing strategies to different regions of the image based on local characteristics. By detecting edge regions and applying selective filtering only in non-edge areas, the method preserves sharp transitions while reducing noise in homogeneous regions, thus eliminating ringing artifacts while maintaining noise reduction benefits
Solution Approach 2:
The filtering approach dynamically adapts to local image characteristics by calculating edge strength metrics and adjusting filtering intensity accordingly. The method transitions between different filtering modes based on local content, applying stronger filtering in smooth regions and no filtering in edge regions, thereby resolving the contradiction between noise reduction and artifact prevention
2Manufacturing precision
If fixed noise reduction filtering is applied to all slices, then processing simplicity is maintained, but effective noise reduction fails due to varying noise levels across different slices with anatomical changes
Solution Approach 1:
The patent performs preliminary classification of slices into homogeneous and non-homogeneous categories based on anatomical content before applying noise reduction. By pre-identifying slices requiring adaptive processing versus those suitable for standard filtering, the method enables effective noise reduction across varying anatomical structures while maintaining computational efficiency through targeted processing
Solution Approach 2:
The method dynamically adjusts filtering parameters based on slice-specific characteristics such as noise level estimates and anatomical content. By modifying filter strength, scale, and type according to each slice's properties, the system achieves effective noise reduction across diverse anatomical regions without requiring complex manual intervention
Data Source
AI summary
The method includes estimating a noise level in a reconstructed image using a total attenuation of the reconstructed image.